Refactor into a base image.

[Alexhuszagh]

No description provided.

[Alexhuszagh]

This still needs to fix deployment and CI: we need to make sure the base image is always built first, and that it is staged for our other image build. It also needs to then be deployed as ghcr.io/cross-rs/base:$tag. The other solution would be to use templates: we run a script which would configure every image, which would avoid deploying another image. But I'm not sure I like that solution.

[Alexhuszagh]

This will probably fail, but I'm wondering to see how the CI orders things so we can use the proper base image.
bors try --target x86_64-unknown-linux-gnu

[bors]

try

Build failed:

• conclusion

[Alexhuszagh]

As expected, this is failing because ghcr.io/cross-rs/ubuntu:main doesn't exist.
https://github.com/cross-rs/cross/actions/runs/3159167477/jobs/5142204364#step:10:95

#3 [internal] load metadata for ghcr.io/cross-rs/ubuntu:main
#3 ERROR: failed to authorize: failed to fetch anonymous token: unexpected status: 403 Forbidden
------
 > [internal] load metadata for ghcr.io/cross-rs/ubuntu:main:
------
Dockerfile.x86_64-unknown-linux-gnu:1
--------------------
   1 | >>> FROM ghcr.io/cross-rs/ubuntu:main
   2 |     ARG DEBIAN_FRONTEND=noninteractive
   3 |     
--------------------
ERROR: failed to solve: failed to fetch anonymous token: unexpected status: 403 Forbidden

We either need to do this incrementally (IE, deploy once then have our builds rely on that, in which case our builds would be out-of-sync with the base images, which would be bad), or need to build and stage the image from the build-base (which I'd need to figure out how to do in CI).

The first option would be bad because if we upgrade our base to say, Ubuntu 22.04, then our first build would use 20.04, not 22.04 since it would use ubuntu:main from the previous build. Any subsequent builds would work after the first merge, but this would be a bad design.

I'm guessing one solution is to save to a tarball and then load from there? I'll look into doing this.

[Emilgardis]

I feel like there should be a GH action already made for saving an image to artifacts and then loading it

[Alexhuszagh]

Will still fail, just testing saving to tarball.
bors try --target x86_64-unknown-linux-gnu

[Emilgardis]

I think I personally prefer having these as

ghcr.io/cross-rs/base:ubuntu-vxxx

less "waste" so to say

[Alexhuszagh]

Our docker load step always needs to load these as main, since that's what we use for our Dockerfiles expect for the base images:

FROM ghcr.io/cross-rs/base:emscripten-main

We technically don't have to deploy these images, but it might be nice for people trying to build new images based off this logic locally, especially cross maintainers.

Our docker load step should also only load the necessary base for our image: centos, ubuntu, emscripten, etc.

[Alexhuszagh]

It seems that the path is the directory for downloading artifacts, not the file name... This should be fixable.

$ ls -la base.tar
drwxr-xr-x  2 runner docker       4096 Oct  4 19:30 .
drwxr-xr-x 15 runner docker       4096 Oct  4 19:30 ..
-rw-r--r--  1 runner docker 1001731072 Oct  4 19:30 base-main-ubuntu.tar

[Alexhuszagh]

bors try --target x86_64-unknown-linux-gnu

[Alexhuszagh]

bors try-
bors try --target x86_64-unknown-linux-gnu

[bors]

try

Build failed:

• conclusion

[Alexhuszagh]

EDIT: It's the --pull flag that's the issue.

Is.... the docker images within the same job not preserved between steps?

https://github.com/cross-rs/cross/actions/runs/3185987473/jobs/5196287429#step:12:14

Load Base Image

Loaded image: ghcr.io/cross-rs/base:main-ubuntu

Build Docker Image

#3 [internal] load metadata for ghcr.io/cross-rs/base:main-ubuntu
#3 ERROR: failed to authorize: failed to fetch anonymous token: unexpected status: 403 Forbidden
------
 > [internal] load metadata for ghcr.io/cross-rs/base:main-ubuntu:
------
Dockerfile.x86_64-unknown-linux-gnu:1
--------------------
   1 | >>> FROM ghcr.io/cross-rs/base:main-ubuntu
   2 |     ARG DEBIAN_FRONTEND=noninteractive
   3 |     
--------------------
ERROR: failed to solve: failed to fetch anonymous token: unexpected status: 403 Forbidden

I guess I can convert load and build to a single step then.

[Emilgardis]

No it 100% should work, otherwise we wouldnt be able to do our tests

This is probably the issue I had with buildx, you need to specify the output

cross/src/docker/custom.rs

Line 157 in 4facf8f

docker_build.args(["--output", "type=docker"]);

[Alexhuszagh]

No it 100% should work, otherwise we wouldnt be able to do our tests

I actually found it, it's the --pull flag where there's no registry entry. Since we never need this in CI, I've added an environment variable to disable it and set it to true in our CI.

[Alexhuszagh]

bors try --target x86_64-unknown-linux-gnu

[bors]

try

Build failed:

• conclusion

[Emilgardis]

thought it could be this #817 (comment)

but maybe not.

[Emilgardis]

We might want to use https://github.com/containers/skopeo#copying-images

from docker/buildx#430, relevant docker/buildx#159 (comment)